Chinese Computer Vision Startups Riding the Auto Wave

Shuai Chen
Jun 22 · 4 min read
(Source: perfectial.com)

Computer vision, among other areas of modern artificial intelligence, is quite fundamental and rather technologically mature. In a little less than a decade, a number of Chinese AI startups have grown enormously into even unicorns with the invasion of computer vision technologies in security & surveillance, banking, mobile payment, etc.

Surveillance was a good entry point for computer vision but perhaps the ultimate application is in autonomous driving.

Before going into details, it is worth clarifying the 3 types of companies in computer vision:

  • Startups like SenseTime, Megvii, Yitu, CloudWalk that quite specialize in algorithms.
  • Traditional surveillance giants supplying end-to-end solutions like Hikvision, Dahua, and Uniview.
  • Tech giants that focus more on back-end software services such as Baidu, Alibaba, and Tencent while Huawei does provide full stack solutions.

Hikvision

Founded in 2001, Hikvision (SZ: 002415) provides video surveillance products and vertical market solutions to global customers. In 2016, a subsidiary was established to focus on intelligent driving technologies. It is involved in R&D, manufacturing and sales of mostly after-market ADAS products to both individual consumers and commercial fleets. The ADAS solutions got certified by 20 OEMs by the end of 2018 and are now applicable to a wider range of brands and models.

Meanwhile, Hikvision is a strategic investor and partner of WHST (Wuhu Sensortech), a millimeter-wave radar specialist. WHST was the first in China to achieve mass production of 77GHz millimeter-wave radar for ADAS applications in 2019. Its radar products for vehicles have been adopted by a number of OEMs by now.

Dahua Technology

Dahua Technology (SHE: 002236) is a video surveillance product and solution provider with worldwide presence. In 2015, Dahua’s Vice Chairman/CTO founded Leapmotor with the company’s investment and technical support.

Leapmotor made №6 among new generation OEMs in terms of sales volume in 2020. Earlier this year, it secured a series B of RMB 4.3 billion with the Hefei government among investors. Leapmotor has always been committed to independent R&D of core technologies in electric vehicles. It has built three vehicle platforms namely T, S and C. It also claims to have developed the electric power system, smart cockpit and ADAS solutions in-house, including AI chips.

SenseTime

SenseTime was founded in 2014 to develop AI technologies and has become a unicorn company. It started exploring AI applications in automotive since 2017 and currently offers two lines of solutions that are SenseAuto Cabin and SenseAuto Pilot.

SenseAuto Cabin includes driver monitoring and analysis for enhanced safety (a.k.a. DMS), facial and gesture recognition for smart cockpit experience, etc. According to SenseTime, it has won over 30 cabin projects with leading auto companies, among which 10+ have been delivered. SensenAuto Pilot brings L2+ to L4 capabilities through in-house developed perception module plus 5 millimeter-wave radars and 1 solid-state LiDAR. The solution will be carried on mass production models by the end of 2022 to early 2023.

Besides software solutions, SenseTime has also developed a mini AV (1/10 the size of a real car) called SenseRover X in 2019, which is for AI lab use.

Orbbec

Founded in 2013, Orbbec is a 3D perception solution provider with various 3D camera products.

In 2018, Orbbec invested in ABAX Sensing, a startup aiming to develop and produce LiDAR system and its core chips for vehicles. Later, ABAX received funding from Bosch as well.

In 2019, Orbbec established a subsidiary company, Oradar, to explore the application of 3D sensing technologies in the cabin. It offers sensor design, module production and algorithm integration to auto companies. The end features include DMS, OMS, facial and gesture recognition, identity verification, etc. Very recently, Oradar released a solid-state LiDAR product for automotive applications that adopted “Addressable VCSEL” and “SPAD” architectures.

ArcSoft

ArcSoft (SHA: 688088) is a computer vision solution provider for mainly smartphones. The company started expanding to automotive and IoT devices in 2018 as the smartphone market was saturating. It offers algorithm-based vision solutions like DMS, OMS, BSD, Interact, Authenticate, AVM, and AR HUD. ArcSoft has achieved rapid growth in the auto sector in 2020 according to its financials. This year, it was reported to have won 37 projects for mass production models and 7 projects for conceptual models.

DeepGlint

DeepGlint is another pioneer in computer vision technologies established in 2013. In 2016, a sister company UISEE was created to focus on autonomous driving by DeepGlint CTO and other industry veterans. After a few years of development, both companies have made progress in commercialization.

DeepGlint provides smart sensors, facial recognitions & behavior pattern analysis, and cloud computing solutions to partners in banking, public and private security sectors.

UISEE is well known by its operation in the Hong Kong International Airport with automated baggage tractors since December 2019. In fact, UISEE solutions have been applied to passenger vehicles, commercial buses and taxis as well as logistics. For example, it won the AVP project for the Baojun E300 model, which belongs to the SAIC-GM-Wuling joint venture company (SGMW). It has also built a fleet of robotaxis in Wuhan with Dongfeng Motor.

Megvii

Founded in 2011, Megvii provides a series of IoT solutions for businesses and the public sector with its computer vision technology and deep learning capability. The company initiated a solution for automotive in 2018. It uses vision-based technologies to recognize owners and unlock vehicles, login for customized cockpit settings, monitor driver behaviors for safety, etc.

Dilusense

Founded in 2015, Dilusense focuses on 3D computer vision technologies by offering modules, sensors or full-stack solutions to end clients. One of the products, 3D cameras, could reach an accuracy of 1 millimeter within 5 meters distance. It performs identification and authentication through the recovery of three-dimensional face information in non-touching manners with finance level of security. The company showcased the 3D camera products at a recent autuonomous driving event aiming to expand the applications in automotive besides banking, smart home and intelligent traffic system.

Geek Culture

Proud to geek out. Follow to join our 1M monthly readers.